How I Run a Performance Marketing Agency With AI Agents (Full Build)

One of the most useful things I've built with AI so far for my company is this agent hub, which has been being used for a lot of different tasks like landing page split tests, copy split tests, ad ideas, and a lot more. And in this video, I want to show you how we built this very specific agent, which is a funnel split test reviewer and a copy agent, which has been built into an automated weekly workflow, or I can just actually manually trigger it myself. So I'll go over the workflow. Why this is important and how I built it. And then I'll also show you some split tests that you could run today inside of your landing pages to increase the conversion rates, whether you're selling a product, selling a service, trying to get more booked calls, or just trying to get leads. I'll show you exactly a few things that will help with that as well. So first things first, just an overview. The trigger is either I can say something, I can send a message to my Claude code session, and I can do that from my phone. Or I can just do it from my computer when I'm working. I also have this on a repeatable schedule every single week, Sunday at noon. It'll go through this workflow and it will do everything. The first thing that it does is actually pulls a lot of the data. So I decided to set this up in a really simple Google Sheet with landing pages for all of my clients or different offers that I'm running. It has the URLs to the landers, the niche, the funnel type, the conversion rate, the priority. And a few other references as well. I also put an ad copy tab that has the audiences, pain points, tones, the offers, and then any kind of like do not do rules. So what happens is the bot that I've trained up on this will go to the Google Sheet, pull all of the relevant data while looking at the instructions that I gave it, and then it'll actually execute. So I'll show you the example of what I built here. This is a document that's got example landing pages, niches, funnel structures, the source of traffic, and then the conversion rate. This is the only column that I update weekly. Every Friday, I'll jump in and change the conversion rate for that past week. So we know if it improved or if it didn't. I have the same thing here set up for the copy tab. We've got the client names, the doc to their copy doc, target audience, core pain point, dream outcome, et cetera, et cetera. And this is a really simple workflow where just an agent inside of Claude Code that I've trained up on certain copywriting and landing page conversion rate optimization principles will actually go through and execute. So you can see here now, once it looks at that data, it'll feed it into 2 agents, a split test reviewer and a copywriting agent. So like I said, the split test reviewer will fetch the client landing page's URL or different offers. Like if you're just doing this for your business, but you've got 5 different offers or 5 different products that you're running, You could have it do it for each one. It'll analyze against a conversion rate optimization framework over 7 dimensions, which I taught it how to do. And I also fed it a ton of winning examples of landing pages. This is super important with anything AI. You have to feed it really, really good info and give it the best prompts and the best resources, and then your outputs will be a lot better. So every week now it's improving. But on top of that, whenever I find a new example of a winner, whenever I find something that I like, whenever a client's landing page is excelling and doing very, very well, I will give it to it so it understands what a winning landing page actually looks like in practice. And it's not just dreaming up something that could be good. So it goes through this agent, split test reviewer, all that kind of good stuff. Then agent 2, it goes through the copywriting agent, very similar kind of a thing. Now I fed it a ton of copy that actually works, examples that work well, and it'll go through and generate different hook angles from an 8-angle rotation. It'll generate 2 full ads per client or per offer. It'll run a kind of quality filter, which this one does as well, and it will go from there. And then it outputs everything into 3 different spots. It outputs this into a ClickUp task for each client, which is really great. We work out of ClickUp. You could do this really in anything. It outputs any of the fresh ad copy to the Google Doc, and then it notifies me in Telegram with a short summary. It tells me what clients were covered, what tests were proposed, what copy was written, and that's all automatically done. Now, this was a really simple build. I built it all using Claude Code, and I basically just told it what I was looking for and what I wanted, and it helped me organize the Google Sheet and the Google Doc to make sure that it could pull the data from it effectively. What I've done is I've made a Claude Code second brain. I cover exactly how to use that inside of the Marketing Mastery Lab school group. You can check it out below right here in AI Mastery. I give you a full overview on how to set up. Your Claude second brain to do stuff just like this. It's a super valuable resource, having some kind of a second brain with AI, and it allows you to build really cool agents like this as well. And if you wanted to, you could deploy them to a VPS or a virtual machine, or you could just keep them running locally, which is totally fine either way. Now that I've got that out of the way, super easy to build, just give it this example if you want something like this done. Let me actually go through a few split tests that I think are super valuable that you should look at doing in your funnel. I'll go over a few different examples here. So I've got 4 funnel types here and 4 different examples of split tests that you should really look at. And these are the same things that my agent is doing and what we do and analyze every single week for both our clients and different offers that we're running internally for our business. So the first one is if you're booking calls, uh, the biggest test that I really recommend for everyone to do is above the fold, that main headline, looking at outcome-driven, versus pain-driven. So you can see up here, this main headline, this is the biggest thing that I would be testing. It's gonna be the highest leverage thing. Anything that people are seeing first is the highest leverage thing that you can test on your page. Next is gonna be testing the call to action. So instead of book a call versus see if you qualify, you could test a few different things here. And lastly, social proof. You could test different avatars or company logos. All this is happening above the fold in the hero section. So you can see right here for the social proof right away, you could either put different pictures of founders, or you could just put different companies. One could work better than the other. Next one is if you're doing a lead magnet or an opt-in funnel. Once again, headline, that's gonna be the biggest, most important thing that you can test most of the time in these kind of tests. Having specificity, so 7 frameworks versus the exact playbook, that could be a really good test. Form fields, testing email only versus email plus first name, et cetera, et cetera. After that, a few more tests can be testing the lead magnet visual itself. So you could do a 3D mockup versus a flat preview. Once again, really easy tests that you can implement if you're running a lead magnet. Next one for a workshop or a webinar funnel, we run a ton of these. These have done really, really well for ourselves and our clients. Same thing, headline curiosity-driven versus outcome-driven, different kinds of urgency. So you could do limited spots versus a countdown timer or both. The big thing when you're doing CTAs is you want people to feel like they're being given something. So reserve my seat or save my spot free can work well. But you can play with that a little bit as well. You can see an example here of a hero section for a page. Now, lastly, e-commerce product pages. So these are going to be really good as well to test different headlines, benefit-led versus product name-led. If you have a really strong, unique value proposition, I think leading with your product name can kind of achieve both of these at once. Price anchoring, super important. Show those savings versus just showing the price. Add to cart, buy now, and ships today. These are super important and not enough people test this. Even in the paid ads manager, when you're doing like learn more versus shop now, the result can be really, really different. And I think testing something like ships today is a huge test that you can do. And we've seen this work really, really well. For a lot of companies that we work with as well. Having a different announcement bar as well can do really, really great. So this is the full landing page split test stack. This is a really strong baseline for things that you should start testing just consistently on a weekly or a biweekly basis. If you want to get access to this, click below, check out the school group. It's free for 7 days. And these examples will be in there that you can reference along a lot of other ones. Hope you enjoyed that. Hope you enjoyed the video. Go out there, make something like this and start seeing better results inside of your paid ads.